Catalyzing net-zero carbon strategies: Enhancing CO2 flux Prediction from underground coal fires using optimized machine learning models

Underground coal fires release substantial carbon dioxide (CO2), posing significant environmental and health threats. Accurate prediction of surface CO2 emissions in these areas is crucial for understanding combustion zones and contributes to the global net zero carbon strategy. Traditional data analysis methods have been inadequate for CO2 flux prediction, highlighting the necessity for advanced machine learning (ML) techniques. This study introduces four optimized ML models-General Regression Neural Networks (GRNNs) and Radial Basis Function Neural Networks (RBFNNs) coupled with Grey Wolf Optimizer (GWO), Particle Swarm Optimization (PSO), and Whale Optimization Algorithm (WOA)-to rapidly predict CO2 flux in areas affected by underground coal fires. Utilizing 223 field test samples, these models consider six key variables: soil temperature at 30 cm depth (ST -30), ambient pressure/temperature/humidity (AP/AT/AH), soil water content (SWC), and wind speed (A -WS). The results underscore the superior predictive accuracy of the GRNN model, with an RMSE of 0.074 and an R2 of 0.9995. Sensitivity analysis reveals A -WS and ST -30 as the most influential factors. Compared to traditional methods, these ML models demonstrate enhanced accuracy and efficiency, marking a significant advancement in the field. The study's findings have broader applications beyond underground coal fires, suggesting potential for these ML models in other environmental monitoring contexts, such as emissions tracking in urban areas or integration with satellite data for global environmental assessment. This methodology represents a pivotal step in environmental management and monitoring, offering scalable and adaptable solutions for various ecological challenges. By rapidly and accurately estimating CO2 flux from underground coal fires, this study contributes significantly to achieving the global net zero carbon target and sets a new benchmark in environmental ML applications.